Your first project

Start here

If glasstarn is not installed yet, do that first: it takes one line in your AI app. See Install.

1. Make a folder for this piece of work, for example Invoice review, and put your export in it. Several files are fine.
2. Open your AI app on that folder. In Codex, Claude Code, Cursor and most others, you start a new project or chat and pick the folder.
3. Say:

Hey, analyze this process.

Your agent finds the export, checks what it needs to understand the data, and opens the results in your browser.

You can also attach a file to a chat that has no folder. That works, but a folder is better. Your export, the results and anything your agent makes for you, such as a summary or slides, stay together in one place. And when you open the same folder next week, your agent picks up where you left off.

No export yet? Open your AI app and say “I want to analyze this process, but I don’t know what to export.” Your agent asks which system the work lives in and tells you exactly what to export. See Don’t know what to export?.

The sections below explain what to expect and how to check the result.

1. Pick one question

If you already know what bothers you, say it. Good first questions are about one kind of item and one thing you care about:

• “Why are our invoices paid late?”
• “Where do support tickets wait the longest?”
• “Which purchase orders go back for approval, and why?”

2. Get an export

You need one file that records what happened to each item, and when. glasstarn needs three things in it:

You need	Example
An ID for each item you follow	Invoice number, order number, ticket ID
What happened	“Invoice received”, “Approved”, “Paid”
When it happened	A date and time

Every other column is welcome: vendor, region, amount, team, who did the step. glasstarn keeps them all and uses them to explain what it finds.

3. What your agent does

Once you ask your question, your agent:

1. Looks at the file and works out which column is the item, which is the step, and which is the time. It tells you what it used, and you can correct it.
2. Checks the data with you. If dates could be read two ways, or some rows look odd, it shows you what it found and asks what to do. Nothing is dropped without your okay. See Data checks.
3. Runs the analysis and gives you a first read: one headline sentence, the top findings with their numbers, what is working well, and two or three questions worth asking next.
4. Saves everything in a glasstarn folder inside your folder, next to your export.

4. Look at the map

When the first read is ready, your agent opens it in your browser by itself and gives you the link. The view opens on the project’s Overview:

• What stands out: your agent’s answer, then the findings, biggest effect first, each with the numbers behind it. Open one to see exactly which items it covers. Findings labelled Driver show which vendors, teams or other fields go together with the slow items. These are patterns, not proven causes.
• Process (in the menu): the map of how work really flows. Thick lines carry the most work and dashed amber shows work going back to an earlier step. Switch to Time and slow hand-offs turn red.

The Read the results section walks through every screen.

5. Keep asking

Ask follow up questions in the same chat: “Is it the same for every region?”, “What happens to the invoices that go back?”, “Which steps could an agent take over?”. The page in your browser keeps up by itself. Ask to see something, such as “show me the invoices over $10,000”, and it moves there with the filter on.

You can close everything and come back another day. Open the same folder in your AI app and ask anything, even just “analyze”. Your agent continues the same project, says so, and picks up the saved questions and findings. Say “start fresh” for a new project. When you get a newer export, put it in the same folder and ask your agent to refresh the project.

Your data stays with you

The export never leaves your computer. Your agent works with the summaries glasstarn calculates, never the full file. See Privacy and trust for exactly what your agent sees.

Next steps

• What data you need
• Read the Overview